Prediction markets for internet points?

post by paulfchristiano · 2019-10-27T19:30:00.898Z · score: 40 (18 votes) · LW · GW · 8 comments


  Why I think this would be good
  Fake accounts and private bets

Using real money in prediction markets is all-but-illegal, and dealing with payments is a pain. But using fake money in prediction markets seems tricky, because by default players have no skin in the game.

Here’s a simple proposal that I think might work reasonably well without being too hard to try:

It may be that a close-enough service already exists, though I’m a bit skeptical.

If you are at all interested in doing this, you have my blessing, this is a pretty generic idea (similar to sites that have existed in the past, e.g. Foresight Exchange). I’d also be happy to pay a bounty for an implementation I think is reasonable, I’d pay at least $1k for something that seems OK and up to $10k for something I think is actually really good.

Why I think this would be good

The barriers to entry in these markets would be tiny compared to existing prediction markets—within 30 seconds of seeing someone mention a market on Facebook, I could have created an account and placed a bet. I would likely bet in this kind of market if a few of my friends did.

Some (though far from all) people would take these markets seriously. For events in the next year or so, I think it’s likely that these markets would quickly converge to better predictions than existing political prediction markets—if prices stayed crazy I think there are plenty of people who would be excited to step in to fix them even if they didn’t take Internet Points very seriously (after all, they might be worth some reputation in the future, the cost is super low, and betting on things is fine). Note that people who step in early to correct mispricings would be able to exit the market after sanity was restored, so wouldn’t even need to tie up their points very long, and that they would have an incentivize to advertise the “easy win” to others after they’d bet in order to free up their capital.

If Internet Points are simple enough for people to really understand, and people expect them to stick around, I think they could potentially capture a big chunk of the altruistic upside from medium-stakes prediction markets.

If the overall workflow is very smooth, I think there is a reasonable chance that the site could spread quite rapidly (spreading just as well to new users as to people who have already made an account).

Note that playing this game can still be a fine experience even if you’ve lost most of your points—going from 250 to 1000 points is no harder than going from 1000 to 4000, it’s just that the numbers are smaller.

Fake accounts and private bets

Ideally you’d have exactly 1 account per person. But creating Facebook accounts is quite cheap. Since every account is granted 1000 points, that’s a problem. Some bad things that could happen, and what you might do about it:

It would be nice to fix these problems in a more robust way, so that market odds are more trustworthy, we can allow private bets, etc. That seems like a big project, over the long run I think the principled solution is basically a credit network: I indicate that I’m willing to trust my friends for IOUs, and then all bets need to be implemented as a string of IOU exchanges. This means that the social network can eventually fragment into different pieces (with the fake accounts basically living in their own part of the social network, which has very few trust connections with real accounts), and the market odds will potentially be different for different people. That makes everything way more subtle to think about, and the implementation seems way more complex, but it seems to me like it should work over the long run and is where you’d really like to be.

(You could also integrate that system with identity verification services who receive a high degree of trust, and then a service itself could effectively be deactivated—have all of its trust exhausted—if it certified a bunch of users who lost money.)


A basic problem with this mechanism is that there is no subsidy (except for noise traders). It’s not as bad as existing prediction markets, where the house takes a huge cut, but not as good as most “real” applications of prediction markets where someone interested in the info is willing to pay for it.

If I’m betting with someone reasonable, at least one of us is losing expected log-points (since the log bakes in risk aversion). So the market can’t work with participants who trust each other’s rationality, unless people just love gambling.

The most natural way to fix this is to subsidize markets. But doing this makes the governance and interpretation of Internet Points harder—does someone with 4000 points have a great history of losing bets, or did they just collect a subsidy? who are we effectively subsidizing, and how do we make that call?—so I’d prefer avoid it. If necessary, probably the nicest way to do it would be for the house to provide a modest amount of liquidity at the prices implied by existing markets. Existing prediction markets are sufficiently irrational that I think this would provide a significant incentive to participate.

If prices on Internet Point markets are used seriously as forecasts, it also introduces an additional set of incentives—some people want to manipulate forecasts (e.g. to make it look like their preferred candidate would be more likely to win the general election). So the average truth-seeking trader expects to get money by betting against them, and this can make the whole system work.

In some sense this is actually the only case where you really needed prediction markets anyway—if everyone is being reasonable then we can just talk it out and share our predictions, and that naive mechanism only really breaks down when some participants believe other participants are being predictably unreasonable (for whatever reason). For better or worse, this is a pretty typical situation.


Comments sorted by top scores.

comment by Scott Shambaugh (scott-shambaugh) · 2019-10-29T20:03:37.343Z · score: 5 (3 votes) · LW(p) · GW(p)

Funny enough, this is *exactly* what is happening over at

comment by Liam Donovan (liam-donovan) · 2019-10-27T20:48:09.153Z · score: 4 (3 votes) · LW(p) · GW(p)
I think it’s likely that these markets would quickly converge to better predictions than existing political prediction markets

Why would you expect this to be true? I (and presumably many others) spend a lot of time researching questions on existing political prediction markets because I can win large sums ($1k+ per question) doing so. I don't see why anyone would have an incentive to put in a similar amount of time to win Internet Points, and as a result I don't see why these markets would outperform existing political prediction markets. Is the idea that many people contributing a minimally-informed opinion will lead to more efficient results than a few people contributing a well-informed opinion + a bunch of noise traders?

comment by Matthew Barnett (matthew-barnett) · 2019-10-28T00:00:02.159Z · score: 1 (1 votes) · LW(p) · GW(p)
($1k+ per question)

I think realistically, the expected value is much lower than $1k per question (on Predictit, for instance), unless you can beat the market by a very substantial margin.

comment by Liam_Donovan · 2019-10-28T10:57:33.821Z · score: 2 (2 votes) · LW(p) · GW(p)

On average, I'd guess my expected PredictIt profit per question I put a substantial time + money investment into is around $600. Obviously, this is evidence that the market is extremely inefficient; I just don't see how a market with no incentive to put the time in would be any better.

comment by John_Maxwell (John_Maxwell_IV) · 2019-10-27T21:07:42.055Z · score: 3 (2 votes) · LW(p) · GW(p)

This reminds me of the ranking system Metaculus has. Maybe they could qualify for the bounty by adding a few features.

comment by Douglas_Knight · 2019-10-30T04:52:32.485Z · score: 2 (1 votes) · LW(p) · GW(p)

What is the problem you are trying to solve and why is this the solution?

It sounds like you are trying to scale up the number of users as fast as possible. Is that the right goal? There are diminishing returns to more users in the presence of a limited set of questions. But scaling the set of questions poses other problems, like adjudicating resolution.

comment by JohnBuridan · 2019-10-27T22:48:25.460Z · score: 1 (1 votes) · LW(p) · GW(p)

For subsidy creation would it work to have a market czar (I mean board of directors) who awards additional points for active participation in questions they are most interested in? I suppose you could also just have a timer which just subsidizes low activity markets to increase their activity, but maybe that would create too many externalities...

comment by Gurkenglas · 2019-10-27T22:16:49.472Z · score: 1 (1 votes) · LW(p) · GW(p)

People could bet that X won't donate dollars to organization Y, and then X can buy points by betting against and donating.